Connect with us

8x8 RGB LED driver.

Discussion in 'Electronic Basics' started by Daniel Pitts, Mar 12, 2013.

Scroll to continue with content
  1. Daniel Pitts

    Daniel Pitts Guest

    TL;DR: I'd like some help with sourcing up-to 340ma from a 1-of-8
    demuxer, and deciding on a 500ma power solution.

    So, coming back to the LED driver design, which is at least a tad more
    complicated than the newbie me thought :) Please pardon my brain-dump
    here. I have some questions near the end.

    My existing design has been posted elsewhere, but lets ignore that and
    see if I can "start from scratch" so-to-speak, and get the right design.

    I have this matrix:
    <http://www.seeedstudio.com/depot/datasheet/2088RGBMatrix.pdf.>

    If I'm reading that right, it is a common anode. I'm not sure about
    whether the "maximum ratings" section is for the entire device, or for
    each LED package. Given the math below, it seems likely to be per LED
    triplet.

    I'll want to change the current value for each color to achieve as pure
    white as possible if all colors are on.

    For now, I'll assume about half as much current for RED, but close to
    equal current for GREEN and BLUE. Lets pick 10ma, 20ma, 20ma for now.

    Given the 2.2v typical drop on Red, and the 3.3v on green/blue, that
    gives me, at full "on", 22mw+66mw+66mw=154mw. Hmm, that is just above
    the spec. Maybe I need to go to 8,16,16?

    So, I can drive one column by a tlc5916 per color.
    <http://www.ti.com/lit/ds/symlink/tlc5917-q1.pdf>

    According to the "Adjusting Output Current" section in the 5917 spec, if
    I want a given output current, Iout, I would use the following formula,
    if the default power-on settings are used:

    Iout = (1.25V/Rext) * 15

    Solving for Rext, I get Rext=(1.25v/Iout) * 15

    This means that for the 8ma I'd use 2343Ω for red and half that, 1.2K,
    for the green/blue. Now, since I don't want to overdrive these, I'd
    probably round up. I'd also probably add a rheostat so I could
    calibrate them more accurately.

    Now, on to the other "side" of the LEDs. There are 8 lines, each one is
    the common anode of 24 LEDS. This means that I would need to source a
    max of (8ma + 16ma + 16ma) * 8 = 320ma. Ouch, that's higher than I'd hoped.

    This is where I need some real help. Are there any 3-to-8 decoder ICs
    that can source that kind of current? If not, is there some sort of 8
    bit buffer IC that I can use, where the input is my 74HC238 decoder, and
    the output can source that current?

    I guess the alternative is to get 8 transistors, and drive them from the
    decoder. Would BJTs be the right device for this job? Would I use NPN,
    or PNP? I guess if I use the 74x238, which is active high, then I'd
    want NPN, 74x138 is actually easier to acquire, so maybe I should
    switch to that and use PNP?

    So, once that is answered, the next part is the power source.
    Apparently I'll need well over 340ma to power this thing. It seems like
    rounding to 500ma would be good enough. I could power it off of a 5v
    500ma AC/DC adapter.

    I was hoping for something more portable though. t looks like if I
    wanted to power it off of AA's, I'd be looking at either 3 cells, and/or
    using a DC-to-DC converter to get 5 volts. The device would run full
    brightness for about 2-3 hours per battery (or 1-2 for rechargable).
    Although, in reality the brightness will be far less than that on
    average. Suggestions on this would be appreciated.

    One of my goals is to prototype this cheaply. I'm not looking to sell
    this design, its mostly just for the experience of designing and
    building it. The digital circuitry is pretty easy and straight forward.
    The analog aspect is still kind of kicking me in the pants.

    Anyway, if you've made it this far, thanks for reading through my
    stream-of-consciousness. Any suggestions (even criticisms) are highly
    appreciated.

    Thanks,
    Daniel.
     
  2. Jon Kirwan

    Jon Kirwan Guest

    Hi, Daniel. I'm still thinking about something similar, too.
    Thoughts below, useful or not:

    I'm thinking that it _should_ be even worse than you imagine.
    I think you are doing a x8 mux for the 8x24 (rgb) matrix, as
    I gather you are using three of the 5916s, one for each
    color.

    Each LED is spec'd at 20mA. That's an average value. The
    absolute max says no more than 70mA peak and 50mA average.
    With a x8 mux, to achieve 20mA average you'd need to drive
    160mA into each. Human intensity perception is logarithmic,
    so shifting to 70mA/8 average from 20mA average will mean
    about 82% brightness, perception-wise. Tolerable. But
    shifting to 20mA/8 drops you to about 65% and 10mA/8 to about
    58%. That's noticeable.

    If you were to peak pulse them at 70mA, you are talking 560mA
    for 8 or 1.68A for all 24 (pushing the red the same.) That's
    a lot more than 340mA. That's assuming you push the red led
    as much as the others, of course.
    I read the absolute max specifications as being "per LED" not
    "per RGB triplet." The 8V reverse spec seems normal on a per
    LED basis. The continuous forward current of 50mA would also
    be "per LED," I think. The peak forward current of 70mA would
    also be "per LED," I think. Because those are quite normal
    for individual LEDs. The power dissipation of 150mW (average,
    of course) makes sense primarily because of the temperature
    range spec. They are allowing for operation of no more than
    60C over ambient. With 150mW, this suggests 400W/C which is
    about twice that of a TO92 BJT (worse than, in short.) Which
    makes sense to me, again on a per LED basis. But only because
    they are telling you that you "could" do that if you only
    operate one of the LEDs. I think that if you run all three of
    them, and they are certainly nearby each other, I'd bet the
    150mW spec would be applied to the triplet, too. So that one,
    I think I'd take as a triplet spec if operating all three.

    Let's say you drive all three at 70mA peak, using a x8 mux
    scheme. Using typical Vf, that's [email protected]/8 + [email protected]/8 +
    [email protected]/8 or 77mW. Even allowing that the voltage is
    "typical" and permitting for another .4V headroom on each
    LED, it's still under 88mW for the triplet. Average. Of
    course, during the 1/8th cycle, the power will be 8 times
    that or slightly over 600mW. But since that is a short time
    and there is a thermal filtering response, it's likely you
    can do that safely. The only concern I'd have is thermal
    flexing of the leds on their mountings within embedded epoxy.
    But I'd go with that for hobby use, anyway.

    Another take on that: They say that the LEDs can support 50mA
    continuous. I think that's "per LED." But... only one of them
    in that case. At 50mA and a typical of 3.3V, you'd see 165mW.
    Which is close to their power dissipation spec.

    The sheet doesn't do a complete analysis for you. It doesn't
    even say much. Nothing about the C/W thermal resistance, for
    example. So it's low on specs and you have to make some
    reasoned guesses. I'm just a hobbyist too, but those are my
    guesses looking at the sheet.

    I'm guessing they are telling you not to drive any LED at
    more than 70mA peak and 50mA average. The operational
    temperature of 85C max is probably the most any particular
    LED should be at. Also, no single LED should dissipate more
    than 150mW on average.
    The rough equivalent for what we did at OSRAM would be to use
    three TLC5916's and use three pots to set the max current.
    Then set up all the LEDs at 25% PWM, turn the whole panel on
    at once, and adjust the three pots until the display read a
    specific (D60) CIE white point using the spectro (and
    software.) Once the pots are set, you are golden. PWM from
    there. The results were quite good enough.

    What I'm tentatively thinking about doing right now is to use
    the TLC5916 on the low side to drive a current mirror on the
    high side... but where I set up the TLC5916 for about 1/10th
    the current. The current mirror will multiply by 10 to
    achieve the actual drive current, using a resistor in one
    leg. That way I can select BJTs quite capable of a
    significant voltage drop AND current and be able to dissipate
    the power without problems. That adds two BJTs per TLC5916
    current sink pin. But that's the price I pay to move the
    dissipation elsewhere while retaining the convenience of a
    shift register and single resistor for setting each current
    sink value.

    But I'll require common cathode arrangement for this, as the
    switches will be on the cathode side, since the current
    mirrors will have to be on the high side.

    I want to pulse as much as 160mA (or a little more) and
    support a voltage drop of more than 2V on the BJT. That's
    20mA = 160mA * 1/8th duty. Can't do that with the 5916 under
    any circumstances.
    That's not going to be white. But you know your experience
    better than I do. Mine says... no guessing... use pots to set
    the white balance. And even then, you may need to calibrate
    each pixel individually. LED manufacturing, 10 years ago at
    least, wasn't up the ability to produce consistent LED
    performance -- even when the LED dies were cut from the exact
    same wafer. In fact, I worked on programming (and some
    optical design) for machines that a big company (Siemens and
    OSRAM) required so they could "bin" their parts before
    selling them, or using them in composite displays. The extra
    expense was not optional. It was required. (Unless the
    customers didn't care and could tolerate "slightly more
    orangish red leds" sitting side by side each other.)

    By the way, red LEDs do not emit much in the green where
    human intensity is perceived. (Only one of the three color
    rods is involved -- usually denoted Y.) I'm not sure why
    you've arbitrarily decided your numbers. But I assume it is
    from personal experience with those matrices. So I just
    caution you, but don't know better than you about them.
    No. Power is AVERAGE. The peak power (ramming Joules into
    something for a short time) does cause some thermal flexing,
    though. And it also creates some excess voltage drops in the
    diodes that doesn't translate directly into light emission.
    So you don't get exactly X times as much light out for X
    times the power in. But the specification is always an
    average specification. If you are pulsing fast, I think the
    thermal response will be very sluggish by contrast and will
    do a good job of "averaging" the pulsed power.
    Yes, I'd definitely use a pot for each of the 5916s.
    As I wrote above, I think it's even worse than you calculate
    if you really want to push these to produce brightness near
    their capacity. Obviously, less works too. But I'd be pushing
    them to do what they can and then use PWM for dimming (and
    geometrically, not linearly.)

    ....

    I'm shooting for 20mA average (I'll never use that much in
    practice), but this means 160mA peak at a x8 mux. In my case,
    it's x5. So just 100mA.

    But I'm also facing limitations in the 5916, in terms of
    dissipation. Aside from the LEDs themselves, all of the
    remaining dissipation takes place in the current sinks. The
    high side switches are ... switches. They don't drop much by
    way of volts. Just a lot of current. And they are external
    BJTs (in my case, anyway.) So they can handle it fine. The
    real problem is with the 5916, which is pathetic at 600mW max
    total. In using all 8 outputs as in your case, this means
    75mW per sink. If I tried to sink 100mA (within spec), that
    would be 750mV. That's all?

    So let's say my LED power rail is 5V and my RED LEDs drop
    2.2V and my high side switch drops 200mV. That leaves 2.6V
    for the sink. Blows that 750mV out of the water! Not even
    close. I'd need a RED LED rail of 3.15V worst case to meet
    the max spec. All because the current sink MUST pick up all
    the excess voltage at the desired current and dissipate it.
    That's what the sink does.

    I could use an external BJT, in common base config, to pick
    up the slack. But then I'd need another power rail at, say
    1.5V to attach to the BJT base. That would move the
    dissipation outside the package and into the BJT. But then
    I'd waste a lot of unnecessary power and voltage headroom
    doing that. Aside from the extra rail.

    I'm considering the idea of a high side current mirror with
    gain (resistor in one leg) and setting up the currents at the
    5916 much lower than I want for the LEDs. The problem there
    is that one of the BJTs will heat up a lot more than the
    other and with Vbe shifting about 2mV/C and perhaps 30C
    difference in temps.. it would be very hard to operate
    correctly without adding feedback on the low side switch so
    that the software could "observe" and correct by changing the
    config registers in the 5916, dynamically.
    So you are looking for high side x8 driver ICs. I see the
    Mitsubishi M54561P, but it's only x7. Not x8. Allegro Micro
    is the site I'd examine.

    http://www.allegromicro.com/en/Prod...-Interface-ICs/High-and-Low-Side-Drivers.aspx
    http://www.allegromicro.com/en/Prod...ICs/High-and-Low-Side-Drivers/UDN2987x-6.aspx

    Their A2982, 2985, and 2987 may work? No experience with
    them.
    Your 5916's are stuck SINKING current. So you need high side
    switches. Even if you drive them with a shift register.
    There's no choice there. You can't use NPN switches with a
    sinking IC.
    Could be even worse, as I indicated earlier. Need to hear
    your response to that, first.
    Keep in mind where all the dissipation takes place. The rail
    voltage(s) are crucial. With a difference between 2.2V and
    3.3V for your LEDs, if you use a SINGLE RAIL for all of them
    then you KNOW in advance that you have to drop an extra 1.1V
    for the RED LEDs and waste that power somewhere. MUCH better
    to use a separate rail there with a switcher supply and not
    have to burn unnecessary power, which only drains your
    batteries that much quicker.
    Yeah. It's frustrating. I'm seeing more clearly just how much
    is really involved in a "good design" of an LED matrix driver
    system. Now I apprehend better why OSRAM used three separate
    external supplies for their modules (switchers were used
    there.) And I also better apprehend why there were 6 ICs in
    there to run a 16x16 grid of RGBs.
    Well, that's my thoughts so far.

    Jon
     
  3. Daniel Pitts

    Daniel Pitts Guest

    Well, that's assuming I want to drive them hard. My experience is that
    they don't need to be near the brightness. If I wanted to make this
    into an out-door direct sunlight display, that'd be different. This is
    more like a desk-top blinky-thing.
    Interesting. I might have to set up a simple circuit to try to drive
    these at their peak pulse with a 1/8th duty cycle at 3.7KHz (which is
    the minimum my application would call for, using 30 FPS). I can afford
    to burn out a matrix for testing purposes. I wouldn't drive it near
    that hard on the real product, since for "debugging" purposes I
    sometimes lower the frequency downward to 1Hz.
    Each triplet, AFAICT, is a single SMD chip. I would think the heat
    dissipation is per triplet too. I wasn't thinking average, as I should
    have been. This means I could in fact drive this thing a lot harder
    than I was thinking, if I need/want to. It's worth an experiment, when
    I have more time.
    The spec even says "Corol" instead of "Color" and a few other Engrishisms.
    I wonder if this would be an issue for outdoor displays in direct
    sunlight. I think it unlikely for my case to get anywhere near those
    extremes.
    Yup, that's kind of what I was thinking, though I'd just eyeball it.
    You can combine the outputs pins to aggregate the current, says so in
    the spec.
    You can if you use two pins instead of 1. You'd have to use twice as
    many 5916s though. Might still be easier than a current mirror. YMMV.
    Meh, close enough for my needs. Anyway, I was mostly picking some
    relative for analysis.
    The pixels in this matrix seems pretty consistent with each other. At
    least, to my eyes.
    Thanks for the tip. It will have to be calibrated more exactly at some
    point, but just trying to do simple analysis.
    Yes, this is true. I wasn't thinking about this, but I'd be willing to
    risk burning out one of these $8 modules to find out you're wrong :)
    The other alternative is that the scaling can be set in "software" by
    sending specific commands to the 5916. That may be better in the end.
    My current circuit just doesn't have any kind of communication to update
    those values without a complete reprogramming, so a physical variable
    resistor is the easiest addition.
    If I wanted true image reproduction certainly. Again, linearly is good
    enough for my purposes.
    Interesting, so you're scanning the segments, and shifting the digits? I
    would have thought you'd mux 7 and shift 5. Or am getting these backwards.
    Hmm, that doesn't quite add up to me. It looks like that max is if
    you're running the thing at ambient 85°C, without heat sinks, where the
    junctions get to be 125°C. I've definitely run it at more than 750mV,
    and it isn't getting even warm. Something isn't quite right with that 750mv.
    I think you're underestimating the 5916. I can't figure out where the
    flaw is in your reasoning, but it seems to me that heat caused by this
    device will be less than your reasoning leads to.
    It depends on orientation doesn't it? As long as I don't exceed the
    maximums on the reverse base/emitter bias.
    Hmm, good point. I'm not yet knowledgeable enough to design such a
    circuit. Hopefully by the end of this Linear Circuits class ;-)
    This is a common Engineers hubris. "This field I've never studied is
    probably easier than everyone else makes it out to be." :) I fall for
    it from time to time.
    Thanks, this does help me set the parameters for my system a little
    better. The 6 ICs figure in the OSRAM module sounds about right,
    assuming they have more output channels than the 8bit ICs I'm using.

    I'm actually considering a non-MCU driven design. I would replace the
    MCU with 1 timer, a counter, 2 RAM chips (for one is the active display,
    the other is the writable buffer) and a some logic ICs to compare the
    values in the RAM to some values in the counter, to handle the PWM aspect.

    After all that, it *might* still be cheaper/easier to go with a cheap
    MCU, but maybe not. Still gotta spec it out a little more.
     
  4. Jon Kirwan

    Jon Kirwan Guest

    Yeah, I understand. I want to design for future flexibility
    so that I don't have to revisit the design every time I
    change something, though. But I also have to find the "sweet
    spot," too. Overdesigning to a point where I'll never care is
    going too far, too.
    I don't think you'll lose it. If you pulse fast, the
    averaging should work out well.
    Let me know what you find out. I think it'll be fine. But
    yeah, I also think the 150mW average has to apply to the
    entire RGB as a unit since that is the basic dissipation
    unit, physically.
    The units I worked on from OSRAM burned as much as 100W per
    module for the 16x16. That's 390mW per RGB or 130mW per LED.
    They were professionally designed for outdoor use in large,
    multi-kilowatt displays. (About 20kW, or so.)

    I'm kind of thinking of running mine at the equivalent of
    about half that, and often much less, per LED. Aggressive,
    but not nearly as aggressive as they were doing. A 5x7, all
    lit up, would be less than 2W. And in normal full brightness
    about a watt. If I dim it down to 25% via PWM (likely, for
    typical use) it's about 1/4 watt average per 5x7. But I'd
    like to design so that I _can_ use them at full brightness in
    certain cases.
    Okay. I'll be using a cobbled up spectro from a DVD-RW and a
    little cardboard box it goes into, with a standard (cheap)
    $10 digital 1Mpixel camera and some software I've already
    written. I'll still have problems because I don't have a good
    way to calibrate pixel intensities over wavelength without
    spending too much money on it. But I'll get by using a cheap
    incandescent (black body radiator) bulb and a standard data
    chart on its emissions to take a hack at it. That is... if I
    ever bother with RGB for anything more than toy use. I may
    not.
    That's true of any current source or sink worth its salt. I'd
    assume it, even if I didn't read it.
    No, that's getting spendy just to increase dissipation. I
    might prefer to bond something to the package, instead,
    though. Or just avoid depending on them for power.
    Yeah. But then let's say you are selling these to an aircraft
    instrumentation maker. They buy a bunch of 7-seg displays to
    use from you. They put them side by side to make up the
    display and someone pays $10,000 for the instrument. It's
    dark at night, while flying, and all the pilot has to do is
    keep staring at the display in the dark. And they see that
    the segments are different colors and brightness and think
    that the instrument maker is putting out cheap stuff and that
    they should pay for a better display. So they don't recommend
    it to their friends. Etc.

    The manufacturers actually DO spend money and time to bin
    their displays because they have to. Not because they want
    to.

    Human perception of shifts in wavelength aren't so good in
    the red -- lousy, actually. But elsewhere our eyes may
    discern even a fraction of a nanometer difference, if next to
    each other. Particularly women, whose wavelength
    discrimination is much better as a rule than men.
    Got it.
    Test away. I'm interested!
    The pots allow me to calibrate the white point externally so
    that the software doesn't have to be recompiled for every
    single display. That's a REAL PAIN. Doing it with the pots
    makes the displays consistent, unit to unit.

    Can you imagine what it would be like to have to carry
    individual calibration information for each and every display
    and to adjust your PWM all over the place to keep track of
    that?
    I want to use the x8 5916 with its ability to individually
    turn on and off specific sinks. That's a nice feature. But
    this means the high side is what is scanned. Has to be.

    My original post about this was doing it the other way. I put
    all the LEDs onto a shared current sink and then turned on
    individual high side switches. But I was forced to use a
    separate BJT for each LED then. Which is what brought on a
    short discussion and led me to look at the 5916, too.
    Now, that's easy to compute. 600mW for the whole package.
    Divide that by 8, you get 75mW. Divide that by the current
    (I'm considering 100mA) and you get 750mV. It's straight
    forward. But you aren't using 100mA. Remember? That's why you
    are able to handle more of a drop in the package. At 20mA,
    for example, it's 5 times more, or 3.75V. So of course you
    don't notice a problem.
    No, my calcs are right, I think.
    Not if I understand your discussion. The 5916 sinks. That's
    what it does. Your matrix has its cathodes worked out so that
    works really well, in fact, using three 5916's there. The
    common anode side is what you need to switch. And you can't
    do that with NPNs (without a still higher voltage rail,
    anyway.) You can drive the PNPs from the low side, of course,
    using an NPN together with each.

    One thing to also keep in mind (or at least I am) is that the
    high side rail might be higher than the microcontroller's.
    For me, this means I can't directly drive a PNP from a micro
    output, since the output of the micro can't "reach" high
    enough to turn the PNP off. In your case, that may not be an
    issue.
    Linear (and others) have boilerplate programs that will
    design them for you. Just plug in what you want and they tell
    you all the parts you need to use.
    Yeah. But you know? It's just an LED!! I mean... cripes. Just
    stick a resistor and you are good, right?

    Then you start thinking about efficiency, parts count, costs,
    power distribution, power dissipation, battery life, size and
    the ability to stack vertically or horizontally, utility and
    cost/benefit to the consumer, brightness, color rendition,
    speed, and specialized software features.... and ...

    Wow. You start looking for an LCD, again. They have fancy
    controller all built in, don't need much power, cost almost
    nothing....

    Of course, there's a reason for LED, too. In my case, LCD
    won't work.
    Hehe. I'll be watching, because frankly I'm interested in
    your thoughts, too.

    Jon
     

  5. <BIG SNIP... of the long winded Jon K.)
    (or is that long fingered? :^)
    I'm not sure it's quite that bad Jon. The 0.6W spec is for an ambient
    temperture of 85C... I assume you can keep it cooler that that.
    Maybe some series PN diodes to move the power dissipation outside of
    the chip? I assume the spec sheet lists the minimum head room voltage
    that it needs to operate the current source. (But I didn't read
    carefully.)

    <more snipping>

    George H.
     
  6. Jon Kirwan

    Jon Kirwan Guest

    But while it's a 40C rise they are looking at, that's on a
    4-layer board with a C/W of 66. On a single layer (or worse,
    as in Daniel's case, a solderless protoboard I think), it's
    103 C/W. About twice as bad. Which is why I'm thinking about
    the 600mW as an absolute limit in my case. Say ambient is
    worst case 45C. Then I've another 40C to work with. But on a
    single layer, I've about twice the thermal resistance, too.
    So I net net.

    Jon
     
  7. Daniel Pitts

    Daniel Pitts Guest

    Ideally, it would be stored in EEPROM in the MCU. Also, the 5196 has a
    calibration register, which adjust the voltage reference across the
    Rext, so you wouldn't have to adjust your PWM at all, just set that
    register on upon boot.

    If this were for production scale, it could be automated to have the
    sensor feed-back into the MCU to set the calibration register. This has
    the benefit of not being a mechanical device.
     
  8. Daniel Pitts

    Daniel Pitts Guest

    Isn't heat transfer exponentially proportional to the difference? If the
    max temp is 125C, and the ambient is 40C, the heat is dissipated much
    faster. Disclaimer, I could be greatly wrong here, this is definitely
    leaving my comfort zone ;-) That's a good thing, makes me think more.
     
  9. Jon Kirwan

    Jon Kirwan Guest

    If this were about EM radiation, it would follow Planck's and
    the integral over wavelength following Stefan-Boltzmann.
    Basically, proportional to T^4 (absolute temp.) If it were
    air conduction, certainly there could be differences.

    But it's not EM radiation that dominates. It's conduction.

    Newton's law (and Fourier's) is something like "the rate of
    heat loss of a body is proportional to the difference in
    temperatures," I think.

    Anyway, that's the conventional use of C/W and it, in
    general, works that way to within a modest degree of error.

    Jon
     
  10. Jon Kirwan

    Jon Kirwan Guest

    Sorry, meant 'air convection'

    Jon
     
  11. Hi Daniel, No it's pretty much linear. You can write a thermal
    equation like ohms law. Heat flux (in Watts) times thermal resistance
    (in degree C/ watt) = temperature difference (in degrees C) Heat flow
    is like electrical current and the temperature difference is like the
    voltage that drives it.

    George H.
     
  12. Daniel Pitts

    Daniel Pitts Guest

    That total makes sense. That does mean that a 125C chip dissipates heat
    approximately twice as fast with air temp being 40C than air temp at 80C.

    So does that mean that with C/W being "66", each watt increases the
    package temperature by 66 degrees (compared to air temp)?
     
  13. Daniel Pitts

    Daniel Pitts Guest

    I did a little reading on Wikipedia (not the most reliable source, I
    know)...

    The 66 C/W implies that each watt requires a difference between the air
    and the junction of 66 degrees per watt. Going back to the 100ma per
    channel (or 800ma total), and a maximum operating temperature of 125C,
    we can work backward to figure out the maximum voltage of the device in
    relation to air temperature.

    (125C - Atemp)/(66 C/W) = Pmax

    They used 85C, which means 40/66 which is 0.6 watts.
    If your air temp is 45C, that would be 80/66, which is 1.2 watts. That
    looks like 1.5 volts max, twice that of Jon's 750mV.

    That is assuming constant 100ma current to all 8 outputs. using the "5"
    outputs, that would be 500ma, which means 2.4volts max. For the Red,
    that means I'd need a LED rail of less than (2.2 + 2.4) 4.6v, and
    green/blue would need to be below (3.3+2.4) 5.7v. It would probably be
    safe to set that at 4v and 5v respectively. Or if you wanted to use more
    common V values, 3.3V for red, and 5.0v for green/blue.

    Jon, I was under the impression your application was monochrome, so you
    could probably get away with a single LED rail.

    For me, I'd be using less current maximum. I'd probably be using closer
    to 20ma per LED, so 160ma total, and my device would be mostly be
    operating indoors at < 40C. This would give me a much larger range for
    voltage, though I'd still probably only use around 5v for both.

    Does this analysis make sense? I mean, it seems to make sense to me, but
    I'm the one making the analysis, and my understanding could be off ;-)

    Thanks,
    Daniel.
     
  14. Jon Kirwan

    Jon Kirwan Guest

    Yes. But keep in mind this is for a 4-layer board!!! Not a
    single layer board. The single layer board says 103 C/W. Not
    66 C/W.
    The reason I came up with 750mV is because it is 103 C/W!!!
    Not 66. I also like a little margin between operation and MAX
    SPEC.
    It's currently mono, though I have some very vague (admitted
    unrealistic for now) idea of doing an RGB some day. Would
    like to design something today that is flexible enough to go
    there someday later.
    Just the 103 C/W vs 66 C/W, mostly. I'm not doing 4-layer
    boards. I'll probably hand solder on vector boards I get from
    Taiwan.

    I'm still stuck with the .75V margin, which I cannot accept.
    So I'm working on an alternative approach.

    Jon
     
  15. Daniel Pitts

    Daniel Pitts Guest

    BTW, I found a newer spec than the one I had linked to, it has more
    detailed thermal information:

    <http://www.ti.com/lit/ds/symlink/tlc5917.pdf>

    I still think your .75v margin is wrong. That's only if you need to
    support 85C still air environment. You can double that margin if you'll
    be running at 45C (which is well more than 100F, if you think in USA
    temps like me).

    Anyway, further reading might be useful to both of us:
    <http://www.ti.com/lit/an/spra953b/spra953b.pdf>

    I think you have a bit more wiggle room than you think.

    Have you actually tried running various currents through your display to
    see how bright each value actually is? I don't know about your specific
    application, but my experience is that 40mA produces an almost painfully
    bright light from the LEDs I'm using.
     
  16. Daniel Pitts

    Daniel Pitts Guest

    Not to mention, a small heatsink with a touch of thermal compound is
    probably all you need to increase your heat dissipation enough to give
    yourself even more wiggle room.

    Look in the new spec above. It depends on which package you have (I have
    PDIP). The PDIP Junction-to-ambient is 51.8 C/W, which is better than
    the old spec's 66 for layer 4 board.

    Anyway, do what you need to do, but I say try a few test runs with the
    thing, and take some actual measurements.

    Good luck!
     
  17. Yup, you've got it!

    "(compared to air temp)?"
    Compared to the ambient temperature around the IC.

    George H.
     
  18. This spec is a little misleading. I'm guessing I could make a one
    layer board work. (mostly ground plane around and under the chip.)
    And certainly two layers with a nice ground plane to take away the
    heat would be the same as 4 layers. Sometimes the thermal
    'resistance' spec will include the area for the ground plane. (That
    was meant for Daniels benefit not yours.)

    George H.
     
  19. Jon Kirwan

    Jon Kirwan Guest

    You make a good point about the resistance spec including
    stuff. It turns out that the newer datasheet shows different
    numbers... as Daniel pointed out yesterday... but also now
    references an entirely different document about changes in
    the way they specify things -- which I still can't say I
    understand. (Which is why I've not responded about it, yet.)

    But anyway, I'll be using vector board.

    Jon
     
  20. Jon Kirwan

    Jon Kirwan Guest

    It's fine, anytime. It's too easy to get mired in one
    thinking mode. Nice to get a kick in a different direction.
    Since it's scanned with very little dead time between
    adjacent scan columns (or rows, depending on how you look at
    it), then from the perspective of the power supply it's a
    nearly continuous high current load. There won't be a lot of
    off period.

    Actually, even that isn't entirely true as the number of
    active LEDs will vary from column to column. But it gets the
    point across that the supply has to handle worst case with
    reasonable results.
    I'm aware, but I don't consider that the tough problem at
    this stage.. for me, anyway.

    Jon
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-